Traffic Incident Duration Prediction: A Systematic Review of Techniques

Abstract

This systematic literature review investigates the application of machine learning (ML) techniques for predicting traffic incident durations, a crucial component of intelligent transportation systems (ITSs) aimed at mitigating congestion and enhancing environmental sustainability. Utilizing the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology, we systematically analyze literature that overviews models for incident duration prediction. Our review identifies that while traditional ML models like XGBoost and Random Forest are prevalent, significant potential exists for advanced methodologies such as bilevel and hybrid frameworks. Key challenges identified include the following: data quality issues, model interpretability, and the complexities associated with high-dimensional datasets. Future research directions proposed include the following: (1) development of data fusion models that integrate heterogeneous datasets of incident reports for enhanced predictive modeling; (2) utilization of natural language processing (NLP) to extract contextual information from textual incident reports; and (3) implementation of advanced ML pipelines that incorporate anomaly detection, hyperparameter optimization, and sophisticated feature selection techniques. The findings underscore the transformative potential of advanced ML methodologies in traffic incident management, contributing to the development of safer, more efficient, and environmentally sustainable transportation systems.